Upload 2 files

.gitattributes

@@ -35,3 +35,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 interview_forge_dataset_FINAL.csv filter=lfs diff=lfs merge=lfs -text
 interview_forge_v3_complete.csv filter=lfs diff=lfs merge=lfs -text
+logo.png filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,270 @@
+import gradio as gr
+import pandas as pd
+import numpy as np
+import os
+import re
+import torch
+from transformers import pipeline
+from sentence_transformers import SentenceTransformer
+from sklearn.metrics.pairwise import cosine_similarity
+
+print("Loading E5 Retrieval Model and Embeddings...")
+
+base_dir = os.path.dirname(__file__)
+csv_path = os.path.join(base_dir, 'interview_forge_v3_complete.csv')
+if not os.path.exists(csv_path):
+    csv_path = os.path.join(base_dir, '..', 'interview_forge_v3_complete.csv')
+
+npy_path = os.path.join(base_dir, 'e5_npu_full_embeddings.npy')
+if not os.path.exists(npy_path):
+    npy_path = os.path.join(base_dir, 'e5_full_embeddings.npy')
+    if not os.path.exists(npy_path):
+        npy_path = os.path.join(base_dir, '..', 'e5_npu_full_embeddings.npy')
+        if not os.path.exists(npy_path):
+            npy_path = os.path.join(base_dir, '..', 'e5_full_embeddings.npy')
+
+df = pd.read_csv(csv_path).dropna(subset=['question']).reset_index(drop=True)
+full_embeddings = np.load(npy_path)
+
+final_model = SentenceTransformer("intfloat/e5-small-v2")
+
+model_id = "Qwen/Qwen2.5-1.5B-Instruct"  
+print(f"Loading {model_id} into memory...")
+generator = pipeline("text-generation", model=model_id, torch_dtype=torch.bfloat16, device="cpu")
+print("Models loaded successfully!")
+
+roles = sorted(df['role'].unique().tolist())
+sectors = sorted(df['sector'].unique().tolist())
+interviewers = ["Strict Technical Lead", "Friendly HR Manager", "Aggressive CISO", "Curious Senior Developer", "Business-Focused Product Manager"]
+levels = ["Junior", "Mid-Level", "Senior", "Expert", "Manager"]
+
+
+def get_interview_question(user_role, user_sector, user_interviewer, user_level):
+    query_text = f"An interview question for a {user_level} {user_role} in the {user_sector} sector, asked by a {user_interviewer}."
+    query_embedding = final_model.encode([f"query: {query_text}"], normalize_embeddings=True)
+    similarities = cosine_similarity(query_embedding, full_embeddings)[0]
+    best_match_idx = similarities.argsort()[::-1][0]
+    return df.iloc[best_match_idx]['question']
+
+
+def get_more_like_this(user_role, user_sector, current_question):
+    if not current_question:
+        return "Please generate a question first."
+    
+    # -------------------------------------------------------------
+    # NEW LOGIC: Filter strictly by the SAME Role and Sector
+    # -------------------------------------------------------------
+    filtered_df = df[(df['role'] == user_role) & (df['sector'] == user_sector)]
+    
+    if filtered_df.empty:
+        filtered_df = df # Fallback if combination is completely missing
+        
+    # Remove the exact question we are looking at so we don't repeat it
+    pool = filtered_df[filtered_df['question'] != current_question]
+    
+    if pool.empty:
+        pool = filtered_df
+        
+    # Pick a random question from this highly relevant pool!
+    random_match = pool.sample(n=1).iloc[0]['question']
+    return random_match
+
+
+def create_circular_progress(grade_text):
+    match = re.search(r'Grade:\s*(\d+)', grade_text)
+    if match:
+        score = int(match.group(1))
+    else:
+        score = 0
+        
+    percentage = (score / 10) * 100
+    dasharray = f"{percentage} {100 - percentage}"
+    
+    if score >= 8:
+        color = "#4ade80" 
+    elif score >= 5:
+        color = "#facc15" 
+    else:
+        color = "#f87171" 
+        
+    svg_html = f"""
+    <div style="display: flex; justify-content: center; align-items: center; padding: 20px; flex-direction: column;">
+        <h3 style="margin-bottom: 15px; color: #cbd5e1; font-family: sans-serif;">AI Grade</h3>
+        <div style="position: relative; width: 150px; height: 150px;">
+            <svg viewBox="0 0 36 36" style="width: 100%; height: 100%;">
+                <path
+                    d="M18 2.0845
+                      a 15.9155 15.9155 0 0 1 0 31.831
+                      a 15.9155 15.9155 0 0 1 0 -31.831"
+                    fill="none"
+                    stroke="#1e293b"
+                    stroke-width="3"
+                />
+                <path
+                    d="M18 2.0845
+                      a 15.9155 15.9155 0 0 1 0 31.831
+                      a 15.9155 15.9155 0 0 1 0 -31.831"
+                    fill="none"
+                    stroke="{color}"
+                    stroke-width="3"
+                    stroke-dasharray="{dasharray}"
+                    style="transition: stroke-dasharray 1s ease-out;"
+                />
+            </svg>
+            <div style="position: absolute; top: 50%; left: 50%; transform: translate(-50%, -50%); font-size: 28px; font-weight: bold; color: #f8fafc; font-family: sans-serif;">
+                {score}/10
+            </div>
+        </div>
+    </div>
+    """
+    return svg_html
+
+def evaluate_and_format(question_text, candidate_answer, user_role, user_sector, user_interviewer, user_level):
+    if not candidate_answer.strip():
+        return "", "Please type an answer before submitting."
+        
+    system_prompt = f"""You are a {user_interviewer} evaluating a {user_level} {user_role} candidate in the {user_sector} sector.
+
+CRITICAL RULES:
+1. Speak DIRECTLY to the candidate using "you" and "your". Never use the word "candidate".
+2. Separate the prompt context. Do NOT penalize or critique the user for constraints that were mentioned in the [INTERVIEW QUESTION].
+3. You MUST generate an Example Answer at the very end. Keep it extremely short.
+
+You MUST output exactly this format and nothing else:
+Grade: [1-10]/10
+Pros:
+- [Pro 1]
+- [Pro 2]
+Cons:
+- [Con 1]
+- [Con 2]
+Example Answer:
+[Provide a strict maximum 2-sentence example of a perfect answer.]"""
+
+    messages = [
+        {"role": "system", "content": system_prompt},
+        {"role": "user", "content": f"[INTERVIEW QUESTION]\n{question_text}\n\n[CANDIDATE'S ANSWER]\n{candidate_answer}"}
+    ]
+    
+    outputs = generator(messages, max_new_tokens=400, temperature=0.7, do_sample=True)
+    raw_feedback = outputs[0]['generated_text'][-1]['content']
+    
+    score_html = create_circular_progress(raw_feedback)
+    clean_feedback = re.sub(r'Grade:.*?\n', '', raw_feedback).strip()
+    
+    return score_html, clean_feedback
+
+# =====================================================================
+# UI DESIGN & THEME INJECTION
+# =====================================================================
+
+custom_css = """
+/* Force the background to be a very dark navy/black to blend with the logo */
+body, .gradio-container {
+    background-color: #040f23 !important;
+}
+/* Center and constrain the width of the dropdown block */
+.centered-dropdowns {
+    max-width: 800px !important;
+    margin: 0 auto !important;
+}
+/* Smaller, centered main generate button */
+.center-btn {
+    max-width: 250px !important;
+    margin: 20px auto !important;
+    display: block !important;
+}
+/* Ensure the question and answer columns are the same height */
+.side-by-side {
+    align-items: stretch !important;
+}
+"""
+
+# Implement a dark, professional theme that utilizes Gold (#d4af37) as an accent color
+theme = gr.themes.Default(
+    primary_hue="amber",
+    secondary_hue="blue",
+    neutral_hue="slate",
+).set(
+    body_background_fill="#040f23",
+    body_text_color="#f8fafc",
+    block_background_fill="#0b172a",
+    block_label_text_color="#cbd5e1",
+    button_primary_background_fill="#d4af37", 
+    button_primary_text_color="#000000",
+    button_secondary_background_fill="#1e293b",
+    button_secondary_text_color="#f8fafc"
+)
+
+with gr.Blocks(theme=theme, css=custom_css) as app:
+    
+    # Render the logo directly using HTML. It will look for "logo.png" in the HF Space directory.
+    gr.HTML(
+        """
+        <div style="text-align: center; margin-bottom: 20px;">
+            <img src="file/logo.png" style="max-height: 250px; margin: 0 auto; display: block;" alt="Interview Forge"/>
+        </div>
+        """
+    )
+    
+    # -------------------------------------------------------------
+    # Centered Dropdown Block
+    # -------------------------------------------------------------
+    with gr.Column(elem_classes="centered-dropdowns"):
+        with gr.Row():
+            role_dropdown = gr.Dropdown(choices=roles, label="Role", value=roles[0] if roles else None)
+            sector_dropdown = gr.Dropdown(choices=sectors, label="Sector", value=sectors[0] if sectors else None)
+        with gr.Row():
+            interviewer_dropdown = gr.Dropdown(choices=interviewers, label="Interviewer Persona", value=interviewers[0])
+            level_dropdown = gr.Dropdown(choices=levels, label="Difficulty Level", value=levels[1])
+            
+    generate_btn = gr.Button("Generate Custom Question", variant="primary", elem_classes="center-btn")
+    
+    gr.Markdown("---")
+    
+    # -------------------------------------------------------------
+    # Side-by-Side Question and Answer block
+    # -------------------------------------------------------------
+    with gr.Row(elem_classes="side-by-side"):
+        with gr.Column():
+            question_display = gr.Textbox(label="The Question", interactive=False, lines=10, text_align="left")
+            more_btn = gr.Button("More Like This", variant="secondary")
+        with gr.Column():
+            user_answer = gr.Textbox(label="Your Answer", lines=10, placeholder="Type your answer here...")
+            submit_btn = gr.Button("Submit Answer for AI Grading", variant="primary")
+            
+    gr.Markdown("---")
+    
+    # -------------------------------------------------------------
+    # Feedback Output
+    # -------------------------------------------------------------
+    with gr.Row():
+        with gr.Column(scale=1, min_width=200):
+            score_circle = gr.HTML()
+        with gr.Column(scale=3):
+            feedback_display = gr.Markdown(label="AI Feedback")
+
+    # -------------------------------------------------------------
+    # Event Wires
+    # -------------------------------------------------------------
+    generate_btn.click(
+        fn=get_interview_question, 
+        inputs=[role_dropdown, sector_dropdown, interviewer_dropdown, level_dropdown], 
+        outputs=question_display
+    )
+    
+    # The new "More Like This" logic passes the Role and Sector as well to filter the dataframe!
+    more_btn.click(
+        fn=get_more_like_this,
+        inputs=[role_dropdown, sector_dropdown, question_display],
+        outputs=question_display
+    )
+    
+    submit_btn.click(
+        fn=evaluate_and_format, 
+        inputs=[question_display, user_answer, role_dropdown, sector_dropdown, interviewer_dropdown, level_dropdown], 
+        outputs=[score_circle, feedback_display]
+    )
+
+if __name__ == "__main__":
+    app.launch(server_name="0.0.0.0", server_port=7860, share=False)